How fast is fast enough? Manufacturers of wireless networking gear tend to have a simple answer: "Whatever we just put on sale." But the picture is always more complicated.

On one hand, WiFi networks send data faster than can any cable or digital-subscriber-line Internet connection, which can make even slow WiFi feel fast to the uninitiated. On the other hand, WiFi never runs as fast as advertised. And so the reality may not come clear until after a long period of tweaking and puzzlement.

We spent two weeks putting three forms of WiFi -- 802.11b, the kind most people use, and two faster versions, 802.11a and 802.11g -- through their paces, testing eight access points and receiver cards.

The reviewed hardware included two 802.11b devices (Netgear's MR814 access point and MA401 card), three 802.11g units (Apple's AirPort Extreme base station and card, and Netgear's WG602 access point), two tri-mode a/b/g components (Linksys's WPC55AG access point and Netgear's WAG511 card) and one dual-mode 802.11a/b access point (D-Link's DI-764).

We set up basic WiFi networks and timed how long each took to transmit a 40.8-megabyte file.

The test environment was a two-story wood-frame house, vintage 1935. Why is this relevant? WiFi signals weaken not just over distance but also as they pass through physical objects such as walls, ceilings and fireplaces. In our tests, those obstructions outweighed distance in attenuating a WiFi signal. When the connection gets sufficiently weak -- in our tests, 20 percent of the original strength -- the WiFi hardware will give up on it.

Our benchmark was a wired network built on 100Mbps Ethernet technology, the traditional way to link computers. That moved the test file in 12 seconds every time -- a speed of about 27 million bits per second.

(Networks, wired or not, are amazingly inefficient for many reasons, including the overhead of their communications protocols. As a rule of thumb, real-world performance of half the promised speed is extraordinarily good, and day-to-day reality is likely to be half of that.)

Standard 802.11b WiFi offers a theoretical peak rate of 11Mbps; with access point and receiving laptop in the same room, the file transfer took slightly less than 80 seconds, for a speed of 4Mbps.

Moving that access point one floor away slowed the transfer time to 107.4 seconds (3.2Mbps). At 25 feet and two walls and a floor away, the signal began to fail. WiFi is supposed to work up to 150 feet away, but without an amplified antenna or open ground between access point and receiver, you will probably never see that range.

(Microwave ovens and many cordless phones can "pollute" WiFi's 2.4GHz frequency, which could also explain some of the fluctuations.)

The two newer technologies, 802.11a and 802.11g, advertise a top speed of 54Mbps but require different trade-offs. The 802.11a technology employs the less-crowded 5.4GHz radio frequency to reduce the odds of interference but doesn't work with other WiFi gear (the D-Link "a" access point got around this incompatibility by adding a separate "b" transmitter). The 802.11g technology uses the same frequency as 802.11b and thus can talk to "b" hardware.

Both new arrivals command a premium over regular WiFi hardware, but "a" costs about twice as much, while "g" tends to cost 50 percent more. The novelty of 802.11g also means that "g" units from different vendors may not cooperate or may need firmware upgrades after the standard is finalized.

The speed improvements both technologies promise are real. In the same room, the 40MB test file zipped across in 23.9 seconds under "a" (13.6 Mbps) and 26.6 seconds under "g" (12 Mbps). Two rooms away, 802.11a ran at 10.4Mbps, but 802.11g had slowed to 6.4Mbps. In the den, where the "b" receivers reached their limit, both "a" and "g" hardware sustained a 4Mbps connection.

If speed alone is a priority, then consider that 802.11a hardware consistently performed about a third better than 802.11g in these tests.

Useful 802.11g connectivity ended about 67 feet, one interior and one exterior wall away, while the "a" network reached about five feet farther. This result confounded our expectations; most reports have found that 802.11a's range falls short of 802.11g's. Interference from other 2.4GHz devices in this and neighboring houses might explain that discrepancy.

The bottom line? The 802.11b technology does fine for most local networking, as long as huge file transfers don't happen often. After all, its real-world speed still outpaces a broadband connection. The 802.11a technology works quite well, but once the 802.11g standard is settled later this year, this technology's better compatibility will make it the obvious choice.

In the meantime, almost anybody can improve the performance of a WiFi network just by rethinking the placement of its access point. Our initial test installation had an access point parked next to the DSL modem in the downstairs office, resulting in sluggish performance upstairs. Whatever version of WiFi we used would have worked better if the hardware had been planted higher up, in a more central part of the house.